Physical geographic regions in the Gulf of California defined using unsupervised learning algorithms

Abstract

The Gulf of California (GOC) is a semi-enclosed sea off northwestern Mexico. It is considered one of the ten biodiversity hotspots of the world and sustains important fisheries at different scales. Unsupervised learning algorithms, particularly clustering techniques, can extract groups or clusters from the raw data; they do not require labelling the data a-priori, thus eliminating the subjectivity when assigning labels manually (i.e., supervised algorithms). In our study, we used a multivariate dataset of physical variables (Sea Surface Temperature, Salinity, Mixed Layer Depth, Sea Surface Height, and the U, V components of the geostrophic surface currents; 0.25° spatial resolution and monthly temporal resolution) and a hierarchical clustering algorithm to define physical regions in the GOC. We also defined minor, intermediate and major regions based on a quantile criterion. Our results indicate that 22 different regions exist in the GOC: eight minor, six intermediate, and eight major. More than one physical region was defined in every previously defined region, suggesting that the surface dynamics of the GOC are more complex than previously described.